Efficient Data Representation Combining with ELM and GNMF

Abstract

Nonnegative Matrix Factorization (NMF) is a powerful data representation method, which has been applied in many applications such as dimension reduction, data clustering etc. As the process of NMF needs huge computation cost, especially when the dimensional of data is large. Thus a ELM feature mapping based NMF is proposed [1], which combined Extreme Learning Machine (ELM) feature mapping with NMF (EFM NMF), can reduce the computational of NMF. However, the random parameter generating based ELM feature mapping is nonlinear. And this will lower the representation ability of the subspace generated by NMF without sufficiently constrains. In order to solve this problem, this chapter propose a novel method named Extreme Learning Machine feature mapping based graph regulated NMF (EFM GNMF), which combines ELM feature mapping with Graph Regularized Nonnegative Matrix Factorization (GNMF). Experiments on the COIL20 image library, the CMU PIE face database and TDT2 corpus show the efficiency of the proposed method.